Spatial pattern of forest resources in a multifunctional landscape
L. Mathysa,b,*, N.E. Zimmermanna, A. Guisanb
a
b
Swiss Federal Research Institute WSL, 8903 Birmensdorf, Switzerland- (lukas.mathys, niklaus.zimmermann)@wsl.ch
University of Lausanne, Dept. of Ecology and Evolution, 1015 Lausanne, Switzerland- antoine.guisan@ie-bsg.unil.ch
KEY WORDS: resource inventory, LIDAR, spatial monitoring, sustainable landscape
ABSTRACT:
Forests are usually assessed as a discrete land cover type. However, we hypothesise that the resources associated with forests show
diverse spatial patterns, only partially matching current forest boundaries. Therefore tree vegetation parameters related to relevant
forest resources at the landscape level were assessed for the Canton of Geneva in western Switzerland. LIDAR based digital terrain
and surface models at the local scale (2.5m resolution) were used to derive the tree vegetation parameters at the landscape scale
(hectare resolution). Resulting rasters of ‘tree vegetation cover’, ‘mean and standard deviation of tree vegetation height’ showed that
areas mapped as forests contained only 53% of the tree vegetation detected from LIDAR. A substantial amount, especially richly
structured and high stands, were found outside forests, located predominantly in suburban areas. LIDAR based assessment of tree
vegetation parameters confirmed that forest resources show a higher diversity in spatial pattern than is implied by the classified forest
area. LIDAR proved to be an efficient technique for large scale forest and landscape assessment and monitoring.
1. INTRODUCTION
Forests provide a diverse range of products and services,
ranging from the original timber production, habitat, protection,
and carbon sequestration to recreation (Farrell, 2000). Each of
these forest resources refers to different aspects of the forested
and often times non-forested land. Hence, they are based on a
specific combination of landscape parameters, or just parameter
segments. In a multifunctional landscape such tree vegetation
parameters relevant to the various forest resources are not
evenly distributed, but show heterogeneous spatial patterns. As
a consequence, also the potential forest resources vary spatially.
Despite this spatial variability of tree vegetation parameters,
forests are usually mapped as discrete land units (Figure 1).
We therefore assessed and analysed the spatial pattern of tree
vegetation parameters, as well as their distribution within a
heterogeneous and multifunctional landscape.
2. MATERIAL AND METHODS
Spatial patterns of forest resources were investigated in the
Canton of Geneva, in western Switzerland (Figure 1).
According to the national mapping agency (Vector 25;
Swisstopo), the study area of 282 km2 included 11% forest and
25% urban areas with a large gradient in landscape
heterogeneity, land use intensity, and urbanisation.
We hypothesize that only a limited number of current forest
resources follow discrete forest borders, while many are
distributed outside or only partially inside forest boundaries.
However, for the sustainable management and monitoring of
forest resources at the landscape level, knowledge of the full
value range within the entire landscape, and not just within a
classified forest, is important (Andersson et al., 2000). To fulfil
this prerequisite for sustainable forest and landscape assessment
and monitoring, the third Swiss National Forest Inventory is
now assessing the full range of raw data values (spatially and
thematically) in its forest assessment using aerial photography.
The resulting point sampling data provides quantitative
estimates. However, the sampled data does not allow for spatial
assessment and monitoring.
The aim of our study is to assess selected tree vegetation
parameters relevant to forest resources for the entire area of
interest, irrespective of a forest boundary.
As many of these parameters represent structural, i.e.,
geometrical aspects of tree vegetation, LIDAR was selected as
the method for raw data acquisition. LIDAR has been used
successfully to assess tree (Morsdorf et al., 2004; Persson,
2002) and stand (Lefsky et al., 2002; Næsset, 2002; Riaño et al.,
2003; Zimble et al., 2003) attributes within forests. However,
tree vegetation parameters must be assessed consistently within
their full value range, i.e. in- and outside forests.
¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
* Corresponding author.
Figure 1. Geneva mapped by discrete land units (Vector 25)
The general approach for the generation of tree vegetation
parameters was to first extract the tree vegetation at the local
level and then derive tree vegetation parameters at the landscape
level.
According to the forest definition of the Swiss National Forest
Inventory tree vegetation has a minimum height of 3 meters
(Keller, 2001). In that approach tree vegetation includes the full
tree canopy, and not just the single point height at the centre of
the tree. We used the difference between a LIDAR-based digital
surface (DSM) and terrain model (DTM) to extract objects taller
- 340 -
International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. XXXVI - 8/W2
than 3m. The raw LIDAR data for the entire study area was
acquired from a helicopter in May 2000 using a small-footprint
LIDAR system. The resulting terrain and surface models had a
grain of 1m, a horizontal accuracy of 30-40cm, and a vertical
accuracy of 25cm. All data processing was performed in
ArcInfo® 8.3.
In a second processing step, tall objects (>3m) other than tree
vegetation were excluded based on the official building map
(accuracy: 10-70cm) and secondary mapping information
(constructed objects in Vector 25). In the building map the
corners of the walls and not the roof are recorded, which is
measured through LIDAR. This resulted in a mean horizontal
discrepancy of 0.8m (n=67) between the polygons of buildings
and constructed objects and the LIDAR raster data. While
cubic-shaped buildings and regular houses matched well with
the raster models, some farm houses with extended roofs
showed larger differences. To account for the horizontal
uncertainty of the DTM/DSM and the polygon maps, we
resampled the elevation models first to a grain of 2.5m using the
bilinear function implemented in ArcInfo® 8.3. The selected
cell size represents a compromise between spatial uncertainty
(cell size > DTM/DSM and map error) and spatial consistency
with other raster data sets.
Some overestimation of tree vegetation occurred around farm
houses with extended roofs (see section 2) and geometrically
complex buildings or construction objects (ex. industrial sites,
overpasses etc.). But compared to the target resolution, i.e.
hectares, this potential bias is negligible. Accuracy of the tree
vegetation height mostly remained within the indicated
uncertainty range. Only some very steep slopes along river
shores resulted in an overestimation of tree vegetation height.
Again, we consider this potential bias as negligible when finally
shifting the focus to the hectare. But we do suggest further
investigations on localised height and tree classification
uncertainties, which might lead to different acquisition densities
depending on the terrain and surface condition.
The assessed tree vegetation was not evenly distributed in the
landscape. While forests cover 11% of the Canton of Geneva,
the assessed tree vegetation based on LIDAR data resulted in a
cover of 19%. Hence, only 53% of the tree vegetation was
found within forest boundaries, 26% in urban and 21% in other
areas. These results support our hypothesis that only a limited
number of forest resources follow the spatial forest pattern (Fig.
2a-c).
In the final processing chain, buildings and constructed objects
were excluded from the resampled DTM/DSM, which had a
difference in height greater than 3m. The cells of the resulting
tree vegetation raster had a size of 2.5m and contained the
height and area values of tree vegetation.
Tree vegetation parameters were then derived at a hectare scale
based on focal analyses with a rectangular moving window of
100m side length applied to the generated 2.5m tree vegetation
rasters. Three key vegetation parameters associated with
different forest resources were generated for this investigation;
‘tree vegetation cover’ as an indicator for horizontal structure
and general tree vegetation quantity (ex. general forest mapping,
timber volume, protection, carbon sequestration), ‘mean height
of tree vegetation’ accounting for biomass (ex. carbon
sequestration, fire risk, habitat), and ‘standard deviation of tree
vegetation height’ accounting for vertical structure (ex. habitat,
recreation). The three resulting tree vegetation rasters covered
the entire canton at a spatial resolution of 1 hectare.
Together, these data sets form a continuous and three
dimensional parameter space, characterising the various key
forest resources. To investigate the spatial relationship between
the parameters, each raster was normalised using a max-min
transformation. The normalised rasters were then stacked to
form an RGB composite.
a)
b)
3. RESULTS AND DISCUSSION
LIDAR based DTM and DSM products were used to assess
forest resource patterns consistently for the entire Canton of
Geneva, i.e. in- and outside official forest boundaries. Relevant
tree vegetation parameters related to forest resources were
assessed at the landscape level based on raw LIDAR data from
the local level. Hence, each hectare cell of the final tree
vegetation parameters was summarised from 1600 cells (2.5m)
containing tree height and area information. At the landscape
level the lower spatial resolution of the resampled DTM/DSM
(2.5m) did not produce significantly different elevation values
compared to the original raw data (1m), but accounted for
spatial uncertainties discussed in section 2. After all, the cell
size of the resampled raw data (2.5m) was still smaller than and
therefore homogenous with respect to the target features, i.e.
tree vegetation, buildings and constructed objects.
- 341 -
c)
Figure 2a-c. Spatial pattern of tree vegetation parameters
International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. XXXVI - 8/W2
In fact, only 89% of the forested area was covered with tree
vegetation. This forest tree vegetation showed highest density in
the study area. However, also urban and more specifically
suburban areas, where tree vegetation represented 20% of the
area, contained high proportions of dense vegetation, providing
much potential for carbon sequestration. On the other hand,
such dense tree vegetation in residential areas also represents a
hazard as a heavy fuel load. Fire danger is not generally high in
Geneva, but the presented approach could also be applied to
Mediterranean areas, where fire is an omnipresent risk.
Forest resources based on structural diversity (ex. habitat,
recreation) were strongly clumped in these suburban areas,
where values of standard deviation of tree vegetation height
were especially high.
Detection of vertical structure, i.e. mean and variance of height,
is a characteristic trait of LIDAR. While fractional tree cover
has been mapped at large scales using spectral data (Fernandes,
2004; Hansen 2003), LIDAR provides additional information
on the height dimension.
The RGB composite of the tree vegetation parameters (Figure
3) confirmed that forests represent resources associated with
dense tree vegetation (ex. timber production, carbon
sequestration). However, considerable forest resources were
found outside the official forest boundaries (ex. carbon
sequestration, fire risk). Especially forest resources related to
structural diversity (ex. habitat, recreation, biodiversity) were
mostly found in these suburban areas outside forests.
REFERENCES
Andersson, F.O., Feger, K.-H., Hüttl, R.F., Kräuchi, N.,
Mattsson, L., Sallnäs, O., Sjöberg, K., 2000. Forest ecosystem
research – priorities for Europe. For. Ecol. Manage., 132(1), pp.
111-119.
Farrell, E.P., Führer, E., Ryan, D., Andersson, F., Hüttl, R.,
Piussi, P., 2000. European forest ecosystems: building the
future on the legacy of the past. For. Ecol. Manage., 132(1), pp.
5-20.
Fernandes, R., Fraser, R., Latifovic, R., Cihlar, J., Beaubien, J.,
Du, Y., 2004. Approaches to fractional land cover and
continuous field mapping: A comparative assessment over the
BOREAS study region. Remote Sens. Environ., 89(2), pp. 234251.
Hansen, M.C., DeFries, R.S., Townshend, J.R.G., Carroll, M.,
Dimiceli, C., Sohlberg, R.A., 2003. Global percent tree cover at
a spatial resolution of 500 meters: First results of the MODIS
vegetation continuous fields algorithm. Earth Interactions,
7(10), pp. 1-15.
Keller, M., 2001. Aerial Photography. In: Swiss National Forest
Inventory: Methods and models of the second assessment,
Birmensdorf, Switzerland, Swiss Federal Research Institute
WSL, pp. 45-64.
Lefsky, M.A., Cohen, W.B., Parker, G.G., Harding, D.J., 2002.
Lidar remote sensing for ecosystem studies. BioScience, 52(1),
pp. 19-30.
Morsdorf, F., Meier, E., Kötz, B., Itten, K.I., Dobbertin, M.,
Allgöwer, B., 2004. LIDAR-based geometric reconstruction of
boreal type forest stands at single tree level for forest and
wildland fire management. Remote Sens. Environ., in press.
Næsset, E., 2002. Predicting forest stand characteristics with
airborne scanning laser using a practical two-stage procedure
and field data. Remote Sens. Environ., 80(1), pp. 88-99.
Persson, Å., Holmgren, J., Söderman, U., 2002. Detecting and
measuring individual trees using an airborne laser scanner.
Photogramm. Eng. Remote Sens., 68(9), pp. 925-932.
Figure 3. Spatial relationship of tree vegetation parameters
represented as an RGB composite
Tree vegetation parameters showed long gradients in suburban
areas. These soft boundaries at the landscape scale represent
rich structured areas with a special microclimatic situation,
therefore positively influencing several forest resources (ex.
habitat, biodiversity, recreation). In contrast, forest borders
contained rather short tree vegetation gradients creating hard
boundaries.
This study showed that some of the traditional forest resources
were found within classified forest boundaries. However,
LIDAR-based landscape assessment also revealed great
potential for further forest resources outside forests.
We therefore conclude that LIDAR data provides an ideal
source for an efficient assessment of forest resources at the
landscape level, therefore enhancing the planning and
sustainable management of multifunctional landscapes.
Riaño, D., Meier, E., Allgöwer, B., Chuvieco, E., Ustin, S.L.,
2003. Modeling airborne laser scanning data for the spatial
generation of critical forest parameters in fire behaviour
modelling. Remote Sens. Environ., 86(2), pp. 177-186.
Zimble, D.A., Evans, D.L., Carlson, G.C., Parker, R.C., Grado,
S.C., Gerard, P.D., 2003. Characterizing vertical forest structure
using small-footprint airborne LiDAR. Remote Sens. Environ.,
87(2-3), pp. 171-182.
ACKNOWLEDGEMENTS
We would like to thank the GIS department of the Canton of
Geneva (SITG) for providing the LIDAR and building data.
DTM, DSM, building data © 2004 by SITG
VECTOR 25 © 2004 swisstopo DV033594
- 342 -